CN108647706B - Article identification classification and flaw detection method based on machine vision - Google Patents
Article identification classification and flaw detection method based on machine vision Download PDFInfo
- Publication number
- CN108647706B CN108647706B CN201810371541.3A CN201810371541A CN108647706B CN 108647706 B CN108647706 B CN 108647706B CN 201810371541 A CN201810371541 A CN 201810371541A CN 108647706 B CN108647706 B CN 108647706B
- Authority
- CN
- China
- Prior art keywords
- image
- picture
- classification
- machine vision
- rotation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention relates to an article identification classification and flaw detection method based on machine vision, which comprises the following steps: identifying and classifying process of the article types based on the support vector machine; the article type identification and classification process comprises a model training process and a model loading and classification realization process; a flaw detection process based on a machine vision flaw detection algorithm; the method comprises the steps of performing similarity comparison on an image correction process and calculating the Hu invariant moment of an image based on a machine vision flaw detection algorithm; and judging whether the image has flaws according to the detection result. The invention can recognize the types of different objects by multi-classifying the training templates and the loading templates of the support vector machine, can realize the detection of different objects, improves the universality of the algorithm and improves the accuracy and the detection speed of flaw detection.
Description
Technical Field
The invention belongs to the field of machine vision detection, and particularly relates to an article identification classification and flaw detection method based on machine vision.
Background
In recent years, with the development of technology, the automatic production of products is one of the main trends of modern production development, and the automatic production has important significance for accelerating the development of social productivity, improving the production technology of enterprises and reducing the labor force of workers.
Under the demand of high-end consumption and rapid increase of capacity, enterprises want to improve competitiveness, and detection of products becomes more and more important. Product surface defects can be divided into a wide variety, for example: wrinkles, scratches, stains, etc. Therefore, flaw detection by machine vision is also essential.
The traditional manual visual detection has low efficiency, high accuracy and dull detection work. Moreover, the industrial production is large, the working time is long, workers are easy to have visual fatigue, and the detection is easy to have the situations of false detection, missed detection and the like. Moreover, the discrimination criteria of manual detection are not quantified, and the quality of the product cannot be guaranteed. Therefore, flaw detection systems based on machine vision have been developed.
In recent years, in the field of machine vision, there are many algorithms for defect detection, but more are applied to surface defect detection of specific articles (such as cloth, carpet, floor tile, etc.), and the versatility is lacking. For example, chinese patent application publication No. CN106872487A, published on 20/6/2017, discloses "a method and apparatus for detecting defects based on visual surface", which, although using a Support Vector Machine (SVM), aims to classify complex texture defects on the surface of a piece of cloth, thereby detecting surface defects of pieces of cloth with different textures, and can be applied to surface defect detection of various pieces of cloth; however, the detection object of the document is still limited to the surface flaw detection of the cloth, and is not applicable to the classification and identification of the articles, and is also not applicable to the flaw detection of different types of articles, and the versatility is not strong.
Disclosure of Invention
In order to solve the problems of the existing flaw detection technology based on machine vision, the invention provides an article identification classification and flaw detection method based on machine vision, which is used for identifying and classifying product images acquired by an industrial camera and carrying out corresponding image processing.
The invention is realized by adopting the following technical scheme: the article identification classification and flaw detection method based on machine vision comprises the following steps:
identifying and classifying process of the article types based on the support vector machine; the article type identification and classification process comprises a model training process and a model loading and classification realization process;
a flaw detection process based on a machine vision flaw detection algorithm; the method comprises the steps of performing similarity comparison on an image correction process and calculating the Hu invariant moment of an image based on a machine vision flaw detection algorithm; and judging whether the image has flaws according to the detection result.
Preferably, the model training process is as follows: firstly, acquiring a training picture downloaded from a picture database in advance, and acquiring training data; then, configuring SVM classifier parameters and training the model; and finally, storing the trained model.
Preferably, the downloaded training picture is positive and negative sample data; the SVM classifier is a linear classifier under the condition of linear divisibility, positive and negative sample data are separated through a decision boundary, and when the distance between the positive and negative sample data and the decision boundary is maximum, the SVM classifier is the best classification selection and finds the maximum classification interval.
Preferably, the configured SVM classifier parameters comprise SVM-ypr type, and optimization is performed by setting the magnitude of the penalty factor C to classify different objects.
Preferably, the loading model implements a classification process as follows: after the model is loaded, the characteristics of the picture collected by the industrial camera are extracted, and the label written in during model training is identified and obtained to obtain an identification result.
Preferably, the figure correction process includes:
(1) firstly, storing a standard template picture in a background;
(2) when an industrial camera acquires a picture of a product to be detected, preprocessing the picture based on an opencv image processing function library, searching a picture contour and storing the picture contour in an iterator;
(3) obtaining vertex coordinates in circulation after the picture contour obtained in the step (2) is obtained;
(4) further screening the area of the outline to obtain an outline graph with a proper size, and if the area of the outline is too small, performing next cycle operation;
(5) according to the screened outline with the proper area, further acquiring the angle between the vertex coordinate and the horizontal line;
(6) acquiring the central point of the contour, and calculating a transformation matrix of rotation and scaling;
(7) and (4) carrying out affine transformation on the picture according to the transformation matrix obtained in the step (6).
In the step (6), the pixel coordinates of the original image are subjected to coordinate transformation for three times: firstly, transforming a coordinate origin from the upper left corner of an image to a rotation center; then, taking the rotation center as an origin, and rotating the image by an angle a; after the rotation is finished, converting the origin of coordinates to the upper left corner of the rotated image;
in the affine transformation process in the step (7), the rotation of the image is divided into two steps: firstly, acquiring a rotation matrix according to a rotation angle and a rotation center; affine transformation is carried out according to the rotation matrix, and the rotation effect of any angle and any center can be realized; and storing the image after the radiation transformation, and finishing the image correction.
According to the technical scheme, the labels set during training are obtained by extracting the features of the target picture, the picture is classified, and the picture is corrected by calling a graph correction algorithm. In the graph correction algorithm, graying and binaryzation are carried out on an image, the vertex coordinates of the graph can be obtained after the image contour is obtained, and the angle between the vertex and the horizontal line can be obtained through the obtained vertex coordinates; and then screening the contour, rotating, calculating a transformation matrix of rotation and scaling, and carrying out affine transformation according to the rotation matrix to obtain a corrected image. Compared with the prior art, the invention has the following beneficial effects:
1. the method adopts a support vector machine to identify and classify the articles, and compares the Hu invariant moment of the target image and the template by the image processing algorithms such as image correction and the like on the acquired image, and has the advantages of high detection universality, higher detection accuracy and higher detection speed. And flaw detection is performed according to a preset standard threshold, so that the detection precision is higher.
2. The detection algorithm adopts a method of a support vector machine to identify and classify different types of articles, so that the purpose that the detection algorithm does not limit the types, shapes, colors and the like of the articles is achieved. The type of defects that can be detected is not limited, and for example, wrinkles, scratches, stains, etc. can be detected.
Drawings
FIG. 1 is a flow chart of a method for machine vision based article identification classification and fault detection according to the present invention;
FIG. 2 is a flow chart of support vector machine based recognition classification;
FIG. 3 is a diagram of a graph rectification flow;
FIG. 4 is a schematic illustration of a pre-image correction;
fig. 5 is a schematic diagram after image rectification.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
Examples
As shown in fig. 1-5, the method for identifying and classifying articles and detecting defects based on machine vision of the present invention can be roughly divided into two steps of identifying and classifying article types and detecting defects.
The method comprises the steps that firstly, an article type identification and classification process is based on a support vector machine and comprises a model training process and a model loading and classification realization process; acquiring training data through positive and negative samples prepared in advance, configuring SVM trainer parameters, training, and storing a trained model; and opening the trained model, then carrying out feature extraction on the detected picture, comparing the data obtained in the training process, extracting matched labels, outputting the types of the labels, and further classifying.
1. Model training process
Firstly, acquiring a training picture downloaded from a picture database in advance, wherein the downloaded training picture is used as positive and negative sample data to acquire training data; then configuring parameters of an SVM trainer (also called SVM classifier) and training the model; and finally, storing the trained model. The SVM classifier can be divided into:
(1) the core idea of a linear classifier in the case of linear classification is the maximum classification interval.
(2) Linear classifiers in the case of linear irreparable introduce the concept of soft intervals.
(3) The nonlinear classifier in the case of linear irreparability is a combination of SVMs and kernel functions.
The configured parameters include:
SVM-ypr type: the support vector class is classified twice, a classification surface is found out, the classification problem is solved, incomplete classification is allowed to be carried out by using an abnormal value punishment factor C, and the purpose of classifying different objects is achieved by setting the size of the punishment factor C for optimization.
kernel _ type: liner, no mapping to high dimensional space, LINEAR discrimination is done in the original feature space, which is the fastest choice. Since items are completed in a pipeline, recognition rate is also an essential consideration. In order to improve the recognition rate, the present embodiment employs a linear kernel type, that is, the SVM classifier employed is a linear classifier in the case of linear classification. The core idea is as follows: a maximum classification interval; positive and negative sample data are separated through a decision boundary, and when the distance between the positive and negative sample data and the decision boundary is the maximum, the method is the best classification selection and finds the maximum classification interval.
2. Loading model to implement classification process
After the model is loaded, the characteristics of the picture collected by the industrial camera are extracted, and the label written in during model training is identified and obtained to obtain an identification result.
And secondly, a flaw detection process, namely a flaw detection algorithm based on machine vision, wherein the algorithm can be roughly divided into a graph correction process and a graph Hu invariant moment calculation process to carry out similarity comparison, and whether flaws exist in the picture is judged according to a detection result. The specific implementation is as follows:
1. process for correcting figures
(1) A standard template picture is stored in the background.
(2) And (5) image preprocessing. When the industrial camera collects the picture of the product to be detected, the picture is preprocessed based on the opencv image processing function library. The preprocessing process comprises graying and binarization processing, and separates an interested target from a background, so that the subsequent processing can be simplified, and the processing speed is increased; and searching the picture contour and storing the picture contour in an iterator, so that the subsequent image correction operation is facilitated.
(3) And (3) obtaining vertex coordinates in a cycle according to the picture contour obtained in the step (2).
(4) And further screening the area of the outline to obtain an outline graph with a proper size, and if the area of the outline is too small, performing the next cycle operation.
(5) And further acquiring the angle between the vertex coordinate and the horizontal line according to the screened outline with the proper area.
(6) The center point of the contour is obtained, and a transformation matrix of rotation plus scaling is calculated.
The pixel coordinates of the original image are subjected to coordinate transformation three times: firstly, transforming a coordinate origin from the upper left corner of an image to a rotation center; then, taking the rotation center as an origin, and rotating the image by an angle a; and after the rotation is finished, converting the origin of coordinates to the upper left corner of the rotated image.
The scaling of the image is mainly used for changing the size of the image, and the width and the height of the scaled image can be changed. To keep the ratio of the image width and height unchanged, it is necessary that the horizontal scaling factor and the vertical scaling factor are equal.
Let the horizontal scaling factor be sx, the vertical scaling factor be sy, (, y0) be the pre-scaling coordinates, and (x, y) be the post-scaling coordinates.
Mapping relationship backwards:
x0＝x/sx；
y0＝y/sy；
when the image is zoomed, firstly, the size of the zoomed image needs to be calculated, wherein newWidth is the width and the height of the zoomed image, and newHeight is the width and the height of the original image. Comprises the following steps:
newWidth＝Sx x width；
newHeight＝Sy x height；
and traversing the zoomed image, and calculating the position of the zoomed pixel in the original image according to the backward mapping relation.
As can be seen from the above formula, the width and height of the scaled image may be multiplied by the original image width and height and the scaling factor.
int rows＝static_cast<int>(src.rows*xRatio+0.5)；int cols＝
static_cast<int>(src.cols*yRatio+0.5)；
The rotation of the image is to rotate the image by a specified angle at a certain point. The rotation process generally uses a cartesian coordinate system with the rotation center as the origin of coordinates, so the first step of image rotation is the transformation of the coordinate system. Let the rotation center be (x0, y0), (x ', y') be the coordinates after rotation, and (x, y) be the coordinates after rotation. The coordinate system is transformed as follows:
x’＝x-x0；
y’＝-y+y0；
the coordinates of the image are then transformed.
x1＝rcosb-a＝rcosbcosa+rsinbsina＝x0cosa+y0sina；
y1＝rsinb-a＝rsinbcosa-rcosbsina＝-x0sina+y0cosa；
The following rotation formula can be obtained: (x ', y') rotated coordinates, (x, y) original coordinates, (x0, y0) center of rotation, angle of a rotation (clockwise).
Backward mapping formula
(7) And (4) carrying out affine transformation on the picture according to the transformation matrix obtained in the step (6).
In the affine transformation process, the specific implementation of the rotation of the image is divided into two steps: firstly, acquiring a rotation matrix according to a rotation angle and a rotation center; and then affine transformation is carried out according to the rotation matrix, so that the rotation effect of any angle and any center can be realized. And storing the image after the radiation transformation, so that the image correction is finished. The image correction aims to improve the detection efficiency and the detection accuracy.
2. Calculating Hu invariant moment of graph and comparing similarity
(1) The method comprises the steps of preprocessing the picture obtained by the image correction, wherein the preprocessing comprises graying processing and binarization processing, and separating an interested target from a background, so that the later processing can be simplified, and the processing speed is increased; the method also comprises an edge treatment which is carried out by adopting a Canny algorithm, and the step comprises four parts: eliminating noise, calculating gradient amplitude and direction, suppressing non-maximum value and delaying threshold value; then searching a picture contour, expressing the contour by using a plurality of vertexes, recording a starting point by using a Freeman chain code, and continuously shifting from the starting point; then, all the outlines are traversed by a tree structure and stored in an array in a recursive mode, and the basic principle is that the edges extracted by the Cannay operator are infinitely approximated.
(2) And comparing the Hu invariant moment of the product picture to be detected with the template picture based on the opencv image processing function library. The moment is an operator for describing image characteristics, and the geometric invariant moment can quickly eliminate the product picture to be detected which is not matched with the template picture, so that the search efficiency is improved. And (2) taking the edge information obtained in the step (1) as input, matching the picture of the product to be detected with the picture of the template, wherein the matching accuracy is higher.
At this time, cvMatchShape functions in the opencv function library are introduced, and the input methods thereof have three matching methods in total:
wherein, in the above formula, hi AIs the Hu moment, h, of object Ai BHu moment of object B.
And comparing the return value of the cvMatchshape function with a previously set standard threshold value, displaying the detection result, and finishing the flaw detection based on the machine vision. The Hu invariant moment has the characteristics of translation, extension, rotation, no deformation and the like. The characteristic quantity composed of the Hu moment is used for identifying the picture, the image identification method has the advantage of high speed, and in order to improve the detection rate of the product, the Hu invariant moment is not more suitable.
The detection algorithm adopts a support vector machine method, trains classifiers through a large number of sample pictures, extracts the characteristics of the target picture, identifies the type of an article, and improves the universality of the algorithm.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (5)
1. The method for identifying, classifying and detecting the flaws of the articles based on the machine vision is characterized by comprising the following steps of:
identifying and classifying process of the article types based on the support vector machine; the article type identification and classification process comprises a model training process and a model loading and classification realization process;
a flaw detection process based on a machine vision flaw detection algorithm; the method comprises the steps of performing similarity comparison on an image correction process and calculating the Hu invariant moment of an image based on a machine vision flaw detection algorithm; judging whether the image has flaws according to the detection result;
the image rectification process comprises:
(1) firstly, storing a standard template picture in a background;
(2) when an industrial camera acquires a picture of a product to be detected, preprocessing the picture based on an opencv image processing function library, searching a picture contour and storing the picture contour in an iterator;
(3) obtaining vertex coordinates in circulation after the picture contour obtained in the step (2) is obtained;
(4) further screening the area of the outline to obtain an outline graph with a proper size, and if the area of the outline is too small, performing next cycle operation;
(5) according to the screened outline with the proper area, further acquiring the angle between the vertex coordinate and the horizontal line;
(6) acquiring the central point of the contour, and calculating a transformation matrix of rotation and scaling;
(7) carrying out affine transformation on the picture according to the transformation matrix obtained in the step (6);
in the step (6), the pixel coordinates of the original image are subjected to coordinate transformation for three times: firstly, transforming a coordinate origin from the upper left corner of an image to a rotation center; then, taking the rotation center as an origin, and rotating the image by an angle a; after the rotation is finished, converting the origin of coordinates to the upper left corner of the rotated image;
in the affine transformation process in the step (7), the rotation of the image is divided into two steps: firstly, acquiring a rotation matrix according to a rotation angle and a rotation center; affine transformation is carried out according to the rotation matrix, and the rotation effect of any angle and any center can be realized; and storing the image after the radiation transformation, and finishing the image correction.
2. The machine vision based article identification, classification and fault detection method of claim 1,
the model training process is as follows: firstly, acquiring a training picture downloaded from a picture database in advance, and acquiring training data; then, configuring SVM classifier parameters and training the model; and finally, storing the trained model.
3. The machine vision based article identification, classification and fault detection method of claim 2,
the downloaded training picture is positive and negative sample data; the SVM classifier is a linear classifier under the condition of linear divisibility, positive and negative sample data are separated through a decision boundary, and when the distance between the positive and negative sample data and the decision boundary is maximum, the SVM classifier is the best classification selection and finds the maximum classification interval.
4. The machine vision based article identification, classification and fault detection method of claim 2,
the configured SVM classifier parameters comprise SVM-ypr types, and optimization is carried out by setting the size of a penalty factor C so as to classify different objects.
5. The machine vision based article identification, classification and fault detection method of claim 1,
the loading model realizes the classification process as follows: after the model is loaded, the characteristics of the picture collected by the industrial camera are extracted, and the label written in during model training is identified and obtained to obtain an identification result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810371541.3A CN108647706B (en) | 2018-04-24 | 2018-04-24 | Article identification classification and flaw detection method based on machine vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810371541.3A CN108647706B (en) | 2018-04-24 | 2018-04-24 | Article identification classification and flaw detection method based on machine vision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108647706A CN108647706A (en) | 2018-10-12 |
| CN108647706B true CN108647706B (en) | 2022-03-22 |
Family
ID=63747135
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810371541.3A Active CN108647706B (en) | 2018-04-24 | 2018-04-24 | Article identification classification and flaw detection method based on machine vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108647706B (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109382968B (en) * | 2018-10-23 | 2021-08-24 | 宁波汇智恒动自动化科技有限公司 | Visual detection and correction system for injection molding machine |
| CN110180799A (en) * | 2019-06-28 | 2019-08-30 | 中船黄埔文冲船舶有限公司 | A kind of part method for sorting and system based on machine vision |
| CN110658215B (en) * | 2019-09-30 | 2022-04-22 | 武汉纺织大学 | PCB automatic splicing detection method and device based on machine vision |
| CN111191723B (en) * | 2019-12-30 | 2023-06-20 | 创新奇智(北京)科技有限公司 | Cascade classifier-based few-sample commodity classification system and classification method |
| TWI783200B (en) * | 2019-12-30 | 2022-11-11 | 緯創資通股份有限公司 | Method and device for determining whether object has defect |
| CN111612737B (en) * | 2020-04-15 | 2023-09-15 | 枞阳立太智能装备有限公司 | Artificial board surface flaw detection device and detection method |
| CN113920075A (en) * | 2021-09-29 | 2022-01-11 | 广州鲁邦通物联网科技股份有限公司 | Simple defect detection method and system based on object identification |
| CN114636618B (en) * | 2022-05-13 | 2022-09-27 | 河南银金达新材料股份有限公司 | PET (polyethylene terephthalate) film stability detection method with stable performance |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106530297A (en) * | 2016-11-11 | 2017-03-22 | 北京睿思奥图智能科技有限公司 | Object grabbing region positioning method based on point cloud registering |
| CN106570511A (en) * | 2016-11-02 | 2017-04-19 | 上海大学 | Keyboard character defect online detection system based on computer vision and keyboard character defect online detection method thereof |
| CN107389701A (en) * | 2017-08-22 | 2017-11-24 | 西北工业大学 | A kind of PCB visual defects automatic checkout system and method based on image |
| CN107730510A (en) * | 2017-09-13 | 2018-02-23 | 电子科技大学 | A kind of combination LTCD and Hu not in the COG of bending moment bump dividing method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6504957B2 (en) * | 1997-07-07 | 2003-01-07 | General Electric Company | Method and apparatus for image registration |
-
2018
- 2018-04-24 CN CN201810371541.3A patent/CN108647706B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106570511A (en) * | 2016-11-02 | 2017-04-19 | 上海大学 | Keyboard character defect online detection system based on computer vision and keyboard character defect online detection method thereof |
| CN106530297A (en) * | 2016-11-11 | 2017-03-22 | 北京睿思奥图智能科技有限公司 | Object grabbing region positioning method based on point cloud registering |
| CN107389701A (en) * | 2017-08-22 | 2017-11-24 | 西北工业大学 | A kind of PCB visual defects automatic checkout system and method based on image |
| CN107730510A (en) * | 2017-09-13 | 2018-02-23 | 电子科技大学 | A kind of combination LTCD and Hu not in the COG of bending moment bump dividing method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108647706A (en) | 2018-10-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108647706B (en) | Article identification classification and flaw detection method based on machine vision | |
| CN110826416B (en) | Bathroom ceramic surface defect detection method and device based on deep learning | |
| CN110314854B (en) | Workpiece detecting and sorting device and method based on visual robot | |
| CN115082419B (en) | Blow-molded luggage production defect detection method | |
| CN111179225B (en) | Test paper surface texture defect detection method based on gray gradient clustering | |
| CN108562589B (en) | Method for detecting surface defects of magnetic circuit material | |
| CN111915704A (en) | Apple hierarchical identification method based on deep learning | |
| CN106651825A (en) | Workpiece positioning and identification method based on image segmentation | |
| CN111178190B (en) | Target detection method and device based on depth image and storage medium | |
| CN109118471A (en) | A kind of polishing workpiece, defect detection method suitable under complex environment | |
| CN110910339B (en) | Logo defect detection method and device | |
| CN112614062A (en) | Bacterial colony counting method and device and computer storage medium | |
| CN114910480A (en) | Wafer surface defect detection method based on machine vision | |
| CN113706490B (en) | Wafer defect detection method | |
| CN108985339A (en) | A kind of supermarket's articles from the storeroom method for identifying and classifying based on target identification Yu KNN algorithm | |
| CN113177924A (en) | Industrial production line product flaw detection method | |
| CN109781737A (en) | A kind of detection method and its detection system of hose surface defect | |
| CN114863464B (en) | Second-order identification method for PID drawing picture information | |
| Niu et al. | Data augmentation on defect detection of sanitary ceramics | |
| CN113516619A (en) | Product surface flaw identification method based on image processing technology | |
| CN111968115B (en) | Method and system for detecting orthopedic consumables based on rasterization image processing method | |
| CN116523916B (en) | Product surface defect detection method and device, electronic equipment and storage medium | |
| CN116433978A (en) | Automatic generation and automatic labeling method and device for high-quality flaw image | |
| CN111707672A (en) | Method for detecting surface defects of wind power rotary supporting piece | |
| CN112184619A (en) | Metal part surface defect detection method based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |